the-tree/block/blk-ioprio.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Block rq-qos policy for assigning an I/O priority class to requests.
0004  *
0005  * Using an rq-qos policy for assigning I/O priority class has two advantages
0006  * over using the ioprio_set() system call:
0007  *
0008  * - This policy is cgroup based so it has all the advantages of cgroups.
0009  * - While ioprio_set() does not affect page cache writeback I/O, this rq-qos
0010  *   controller affects page cache writeback I/O for filesystems that support
0011  *   assiociating a cgroup with writeback I/O. See also
0012  *   Documentation/admin-guide/cgroup-v2.rst.
0013  */
0014
0015 #include <linux/blk-mq.h>
0016 #include <linux/blk_types.h>
0017 #include <linux/kernel.h>
0018 #include <linux/module.h>
0019 #include "blk-cgroup.h"
0020 #include "blk-ioprio.h"
0021 #include "blk-rq-qos.h"
0022
0023 /**
0024  * enum prio_policy - I/O priority class policy.
0025  * @POLICY_NO_CHANGE: (default) do not modify the I/O priority class.
0026  * @POLICY_NONE_TO_RT: modify IOPRIO_CLASS_NONE into IOPRIO_CLASS_RT.
0027  * @POLICY_RESTRICT_TO_BE: modify IOPRIO_CLASS_NONE and IOPRIO_CLASS_RT into
0028  *      IOPRIO_CLASS_BE.
0029  * @POLICY_ALL_TO_IDLE: change the I/O priority class into IOPRIO_CLASS_IDLE.
0030  *
0031  * See also <linux/ioprio.h>.
0032  */
0033 enum prio_policy {
0034     POLICY_NO_CHANGE    = 0,
0035     POLICY_NONE_TO_RT   = 1,
0036     POLICY_RESTRICT_TO_BE   = 2,
0037     POLICY_ALL_TO_IDLE  = 3,
0038 };
0039
0040 static const char *policy_name[] = {
0041     [POLICY_NO_CHANGE]  = "no-change",
0042     [POLICY_NONE_TO_RT] = "none-to-rt",
0043     [POLICY_RESTRICT_TO_BE] = "restrict-to-be",
0044     [POLICY_ALL_TO_IDLE]    = "idle",
0045 };
0046
0047 static struct blkcg_policy ioprio_policy;
0048
0049 /**
0050  * struct ioprio_blkg - Per (cgroup, request queue) data.
0051  * @pd: blkg_policy_data structure.
0052  */
0053 struct ioprio_blkg {
0054     struct blkg_policy_data pd;
0055 };
0056
0057 /**
0058  * struct ioprio_blkcg - Per cgroup data.
0059  * @cpd: blkcg_policy_data structure.
0060  * @prio_policy: One of the IOPRIO_CLASS_* values. See also <linux/ioprio.h>.
0061  */
0062 struct ioprio_blkcg {
0063     struct blkcg_policy_data cpd;
0064     enum prio_policy     prio_policy;
0065 };
0066
0067 static inline struct ioprio_blkg *pd_to_ioprio(struct blkg_policy_data *pd)
0068 {
0069     return pd ? container_of(pd, struct ioprio_blkg, pd) : NULL;
0070 }
0071
0072 static struct ioprio_blkcg *blkcg_to_ioprio_blkcg(struct blkcg *blkcg)
0073 {
0074     return container_of(blkcg_to_cpd(blkcg, &ioprio_policy),
0075                 struct ioprio_blkcg, cpd);
0076 }
0077
0078 static struct ioprio_blkcg *
0079 ioprio_blkcg_from_css(struct cgroup_subsys_state *css)
0080 {
0081     return blkcg_to_ioprio_blkcg(css_to_blkcg(css));
0082 }
0083
0084 static struct ioprio_blkcg *ioprio_blkcg_from_bio(struct bio *bio)
0085 {
0086     struct blkg_policy_data *pd = blkg_to_pd(bio->bi_blkg, &ioprio_policy);
0087
0088     if (!pd)
0089         return NULL;
0090
0091     return blkcg_to_ioprio_blkcg(pd->blkg->blkcg);
0092 }
0093
0094 static int ioprio_show_prio_policy(struct seq_file *sf, void *v)
0095 {
0096     struct ioprio_blkcg *blkcg = ioprio_blkcg_from_css(seq_css(sf));
0097
0098     seq_printf(sf, "%s\n", policy_name[blkcg->prio_policy]);
0099     return 0;
0100 }
0101
0102 static ssize_t ioprio_set_prio_policy(struct kernfs_open_file *of, char *buf,
0103                       size_t nbytes, loff_t off)
0104 {
0105     struct ioprio_blkcg *blkcg = ioprio_blkcg_from_css(of_css(of));
0106     int ret;
0107
0108     if (off != 0)
0109         return -EIO;
0110     /* kernfs_fop_write_iter() terminates 'buf' with '\0'. */
0111     ret = sysfs_match_string(policy_name, buf);
0112     if (ret < 0)
0113         return ret;
0114     blkcg->prio_policy = ret;
0115     return nbytes;
0116 }
0117
0118 static struct blkg_policy_data *
0119 ioprio_alloc_pd(gfp_t gfp, struct request_queue *q, struct blkcg *blkcg)
0120 {
0121     struct ioprio_blkg *ioprio_blkg;
0122
0123     ioprio_blkg = kzalloc(sizeof(*ioprio_blkg), gfp);
0124     if (!ioprio_blkg)
0125         return NULL;
0126
0127     return &ioprio_blkg->pd;
0128 }
0129
0130 static void ioprio_free_pd(struct blkg_policy_data *pd)
0131 {
0132     struct ioprio_blkg *ioprio_blkg = pd_to_ioprio(pd);
0133
0134     kfree(ioprio_blkg);
0135 }
0136
0137 static struct blkcg_policy_data *ioprio_alloc_cpd(gfp_t gfp)
0138 {
0139     struct ioprio_blkcg *blkcg;
0140
0141     blkcg = kzalloc(sizeof(*blkcg), gfp);
0142     if (!blkcg)
0143         return NULL;
0144     blkcg->prio_policy = POLICY_NO_CHANGE;
0145     return &blkcg->cpd;
0146 }
0147
0148 static void ioprio_free_cpd(struct blkcg_policy_data *cpd)
0149 {
0150     struct ioprio_blkcg *blkcg = container_of(cpd, typeof(*blkcg), cpd);
0151
0152     kfree(blkcg);
0153 }
0154
0155 #define IOPRIO_ATTRS                        \
0156     {                           \
0157         .name       = "prio.class",         \
0158         .seq_show   = ioprio_show_prio_policy,  \
0159         .write      = ioprio_set_prio_policy,   \
0160     },                          \
0161     { } /* sentinel */
0162
0163 /* cgroup v2 attributes */
0164 static struct cftype ioprio_files[] = {
0165     IOPRIO_ATTRS
0166 };
0167
0168 /* cgroup v1 attributes */
0169 static struct cftype ioprio_legacy_files[] = {
0170     IOPRIO_ATTRS
0171 };
0172
0173 static struct blkcg_policy ioprio_policy = {
0174     .dfl_cftypes    = ioprio_files,
0175     .legacy_cftypes = ioprio_legacy_files,
0176
0177     .cpd_alloc_fn   = ioprio_alloc_cpd,
0178     .cpd_free_fn    = ioprio_free_cpd,
0179
0180     .pd_alloc_fn    = ioprio_alloc_pd,
0181     .pd_free_fn = ioprio_free_pd,
0182 };
0183
0184 void blkcg_set_ioprio(struct bio *bio)
0185 {
0186     struct ioprio_blkcg *blkcg = ioprio_blkcg_from_bio(bio);
0187     u16 prio;
0188
0189     if (!blkcg || blkcg->prio_policy == POLICY_NO_CHANGE)
0190         return;
0191
0192     /*
0193      * Except for IOPRIO_CLASS_NONE, higher I/O priority numbers
0194      * correspond to a lower priority. Hence, the max_t() below selects
0195      * the lower priority of bi_ioprio and the cgroup I/O priority class.
0196      * If the bio I/O priority equals IOPRIO_CLASS_NONE, the cgroup I/O
0197      * priority is assigned to the bio.
0198      */
0199     prio = max_t(u16, bio->bi_ioprio,
0200             IOPRIO_PRIO_VALUE(blkcg->prio_policy, 0));
0201     if (prio > bio->bi_ioprio)
0202         bio->bi_ioprio = prio;
0203 }
0204
0205 void blk_ioprio_exit(struct request_queue *q)
0206 {
0207     blkcg_deactivate_policy(q, &ioprio_policy);
0208 }
0209
0210 int blk_ioprio_init(struct request_queue *q)
0211 {
0212     return blkcg_activate_policy(q, &ioprio_policy);
0213 }
0214
0215 static int __init ioprio_init(void)
0216 {
0217     return blkcg_policy_register(&ioprio_policy);
0218 }
0219
0220 static void __exit ioprio_exit(void)
0221 {
0222     blkcg_policy_unregister(&ioprio_policy);
0223 }
0224
0225 module_init(ioprio_init);
0226 module_exit(ioprio_exit);